prithivMLmods/Zenith-9B-CodeCore-Merge

VISIONPricing:Input $0.431 / Cached $0.0862 / Output $1.12Concurrent Unit Cost:1Model Size:9BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Sep 15, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Featherless Exclusive Cold

prithivMLmods/Zenith-9B-CodeCore-Merge is a 9-billion parameter merged language model, built upon Qwen3.5-9B and integrated with OxCoder-9B, Qwopus3.5-9B-Coder, and Ornith-1.5-9B. Designed for long-horizon coding tasks, it excels in agentic coding, multi-step problem solving, and complex software engineering workflows. This model specializes in sustained reasoning, code understanding, modification, and debugging within a 32768-token context window.

Loading preview...

Zenith-9B-CodeCore-Merge Overview

Zenith-9B-CodeCore-Merge is a 9-billion parameter model developed by prithivMLmods, specifically engineered for advanced coding and reasoning tasks. It is a merge of several specialized models, using Qwen3.5-9B as its base, combined with OxCoder-9B, Qwopus3.5-9B-Coder, and Ornith-1.5-9B.

Key Capabilities

  • Long-horizon coding: Designed to handle complex, multi-step coding projects.
  • Agentic coding and reasoning: Optimized for autonomous coding workflows, including code understanding, modification, and debugging.
  • Multi-step problem solving: Combines the strengths of its constituent models for sustained reasoning across various tasks.
  • Instruction following: Enhanced for precise execution of complex instructions in software engineering contexts.

Good For

This model is intended for developers and researchers working on:

  • Complex software engineering tasks requiring deep code understanding.
  • Automated code generation, refactoring, and debugging.
  • Agentic systems that need to perform multi-step coding operations.
  • Projects demanding sustained reasoning and problem-solving in a coding environment.

It's important to note that this model is experimental and may produce artifacts. A GGUF version is available, though it does not preserve the Multi-Token Prediction (MTP) heads and operates as a standard autoregressive decoder.